Skip to content

GitHub-Hosted and Self-Hosted Runners

Overview

Decide when GitHub-hosted runners are enough, when self-hosted capacity is required, and how labels, runner groups, and autoscaling keep jobs on the right fleet.

A runner is the machine that executes a job. GitHub-hosted runners are ephemeral VMs (or containers) GitHub operates — labelled ubuntu-latest, windows-latest, macos-latest, plus larger and GPU SKUs on paid plans. Self-hosted runners are machines you register; they can reach private networks, use custom images, and scale under your cloud account. Runner groups and labels control which repositories may use which capacity.

This is a core tutorial in Module 3 · Runners of the REBASH Academy GitHub Actions for Cloud & DevOps Engineers series — written for Cloud, DevOps, Platform, and SRE engineers.

Prerequisites

Learning Objectives

By the end of this tutorial, you will be able to:

  • Contrast GitHub-hosted and self-hosted trade-offs
  • Select runs-on labels intentionally
  • Explain organisation runner groups and repository access
  • Outline why autoscaling exists (queue depth and cost)
  • List security risks of unprotected self-hosted runners

Architecture

This topic’s control points and relationships are shown below.

Runner architecture

Theory

What it is

GitHub schedules jobs; runners claim them. With hosted runners you only declare runs-on: ubuntu-latest (or a specific image / larger runner label). GitHub provisions a clean environment, runs your steps, and destroys the VM afterward — strong isolation for public pull requests.

Self-hosted runners install the Actions runner agent on a VM, bare metal, or Kubernetes-backed pool. You choose the operating system, tools, and network path (for example, access to an internal Artifactory or a private Kubernetes API). Registration associates the runner with a repository, organisation, or enterprise. Labels (default OS labels plus custom ones such as gpu or prod-build) appear in runs-on: so only matching jobs land on that machine. Runner groups (organisation / enterprise) restrict which repositories may target a set of runners — essential when privileged builders must not be usable by every forkable repo.

Autoscaling means capacity grows when the pending job queue is deep and shrinks when idle. Patterns include cloud VM scale sets, Actions Runner Controller (ARC) on Kubernetes, and vendor autoscalers. Cold starts trade cost savings for slower first jobs; warm pools mitigate that.

Kind Who operates Typical use
GitHub-hosted GitHub Public CI, standard Linux/Windows/macOS
Larger / GPU hosted GitHub (paid) Heavier builds without self-hosting
Self-hosted You Private network, custom hardware, compliance

Why it matters

Wrong runner choices create security and reliability debt: a self-hosted runner on a developer laptop that accepts jobs from a public fork is a classic compromise path; missing labels leave GPU jobs queued forever; static fleets burn money overnight. Platform teams treat runners as product infrastructure — patched, monitored, labelled, and grouped like any other fleet. Minute quotas on hosted runners also drive FinOps conversations: cache well, fail fast, and move long private builds to self-hosted when economics favour it.

How it works

  1. Job becomes ready; GitHub looks for a runner matching runs-on labels and access scope.
  2. Hosted: assign an ephemeral VM/image with preinstalled software (see runner image notes).
  3. Self-hosted: idle agent polls; on claim, it executes steps in the configured work directory.
  4. Logs stream to GitHub; the job concludes; hosted VMs are discarded; self-hosted agents return to idle (unless ephemeral mode tears them down).
  5. Autoscalers watch queue metrics or webhook signals and create or delete runner instances.

You can complete this module’s YAML without registering a self-hosted runner — use ubuntu-latest and document when you would switch. Register a self-hosted agent only on machines you control, never on untrusted shared desktops for public repositories.

Key concepts and comparisons

Concern GitHub-hosted Self-hosted
Isolation for forks Strong (ephemeral) You must harden / use ephemeral runners
Private network Limited Native
Custom tools / GPUs Larger SKUs or workaround Full control
Ops burden Low Patching, scaling, secrets on host
Cost model Minutes Cloud VM + idle waste
Control Purpose
Labels Match job requirements to capacity
Runner groups Limit which repos can use privileged fleets
Ephemeral self-hosted One job per instance — closer to hosted isolation

Common pitfalls

  • Registering a personal laptop as an organisation runner for open-source forks.
  • Using self-hosted runners for untrusted pull_request workflows from forks without isolation.
  • Forgetting custom labels so jobs never start (or omitting labels so the wrong machine takes them).
  • Assuming self-hosted disks are clean between jobs — leftover files and Docker images cause flaky builds unless you use ephemeral runners or strict cleanup.
  • Equating “Docker available on the runner” with safe Docker-in-Docker for every workload.

Hands-on Lab

Create a workspace for this tutorial.

mkdir -p ~/rebash-github-actions/module-03/.github/workflows && cd ~/rebash-github-actions/module-03/.github/workflows

Focus: contrast github-hosted labels with a self-hosted tagged job

Step 1 – Hosted vs self-hosted definitions

mkdir -p .github/workflows notes
cat > notes/runners.md << 'EOF'
# Runner notes
- github-hosted: ubuntu-latest — patched by GitHub, ephemeral
- self-hosted: you patch OS, scale, isolate secrets
- Prefer labels like [self-hosted, linux, rebash]
EOF
cat > .github/workflows/runners.yml << 'EOF'
name: Runner shapes
on: [workflow_dispatch]
permissions:
  contents: read
jobs:
  hosted:
    runs-on: ubuntu-latest
    steps:
      - run: uname -a
  self_hosted_shape:
    runs-on: [self-hosted, linux, rebash]
    if: false
    steps:
      - run: echo "Enable when a labelled runner exists"
EOF

Step 2 – Validate labels

grep -E 'ubuntu-latest|self-hosted' .github/workflows/runners.yml
test -f notes/runners.md

Final step – Cleanup note

# Keep ~/rebash-github-actions/ for later tutorials

Validation

  • Lab commands run under ~/rebash-github-actions/module-03/.github/workflows/
  • You can explain each Theory section in your own words
  • You used modern tooling where it applies to this topic
  • You can describe one production failure mode for this topic

Code Walkthrough

Production practice for GitHub-Hosted and Self-Hosted Runners always combines:

  1. Inspect before you change (status, plan, logs, dry-run)
  2. Prefer reversible, documented changes (Git, IaC, drop-ins, version pins)
  3. Capture evidence (command output, pipeline logs) for handovers
  4. Prefer current tools and APIs over legacy shortcuts
  5. Least privilege — escalate credentials only when required

Keep runbooks short enough to follow under pressure. Automate checks; keep humans for judgement.

Security Considerations

  • Treat credentials and tokens for github-actions as privileged — never commit them
  • Prefer short-lived auth (OIDC, roles, SSO) over long-lived keys
  • Validate blast radius before apply/deploy/delete operations
  • Restrict who can approve production changes
  • Collect audit logs; limit who can read sensitive traces

Common Mistakes

Registering a personal laptop as an organisation runner for open-source forks.

Validate assumptions against the Theory section and official docs before changing production.

Using self-hosted runners for untrusted pull_request workflows from forks without isolat

Lab shortcuts (open security groups, admin roles, skip approvals) must not ship unchanged.

Changing production without a rollback path

Always know how to revert (previous artefact, prior release, state rollback, DNS failback).

Best Practices

  • Encode GitHub-Hosted and Self-Hosted Runners changes as code and review them in pull requests
  • Pin versions (images, modules, actions, provider plugins)
  • Separate environments with clear promotion gates
  • Alert on symptoms with runbooks attached
  • Destroy lab resources; tag everything with owner and expiry where possible

Troubleshooting

Symptom Likely cause Fix
Auth / permission denied Wrong identity, policy, or scope Check caller identity, roles, and least-privilege policies
Timeout / no route Network, DNS, security group, or endpoint Trace path, DNS, and allow-lists before retrying
Drift / unexpected plan Manual change or wrong state/workspace Reconcile desired vs actual; avoid click-ops on managed resources
Pipeline/job red Flaky step, cache, or missing secret Read failing step logs; bisect recent workflow/config changes
Cost spike Idle load balancer, NAT, oversized compute Inventory billable resources; stop/delete labs promptly

Summary

GitHub-Hosted and Self-Hosted Runners is essential for Cloud and DevOps engineers working with github-actions. Practise the lab until the inspection and change path is muscle memory, then continue the track.

Interview Questions

  1. Pros and cons of github-hosted versus self-hosted runners?
  2. Jobs queued forever on self-hosted — what do you check?
  3. Why are labels important for runner pools?
  4. What isolation risks do self-hosted runners have with public forks?
  5. How do you patch and monitor self-hosted fleets?

Sample answer — question 2

Verify the runner is online, labels match runs-on, and the repository is allowed to use the runner group.

Sample answer — question 4

Never use self-hosted runners for untrusted public fork PRs without hard isolation.

References